Proton exchange membrane fuel cell life prediction method based on deep learning

Abstract

The invention discloses a method for predicting the service life of a proton exchange membrane fuel cell based on deep learning. The method comprises the following steps: carrying out noise reduction processing on proton exchange membrane fuel cell data by using discrete wavelet transform, and extracting feature data; the characteristic data are input into a trained deep learning model, predicted voltage data of the proton exchange membrane fuel cell are obtained, then the remaining useful life of the cell is determined based on the predicted voltage data, and the deep learning model is constructed based on a Non-station Transformer model. The method can effectively improve the prediction precision of the remaining useful life of the cell, predicts the attenuation state of the proton exchange membrane fuel cell, and facilitates the design of a related energy management strategy to prolong the service life of the cell.

Description

Technical Field

[0001] The present invention relates to the field of fuel cell management technology, and more specifically, to a proton exchange membrane fuel cell life prediction method based on deep learning. Background Art

[0002] Hydrogen energy is a secondary energy source that is abundant in source, green, low-carbon and widely used. At present, new energy vehicles are developing rapidly, and hydrogen fuel cell vehicles have attracted widespread attention due to their advantages such as high efficiency and cleanliness. Proton Exchange Membrane Fuel Cell (PEMFC) converts the chemical energy contained in hydrogen into electrical energy, and has the advantages of cleanliness, high power density and low operating noise. PEMFC is a nonlinear complex system with multi-physical fields and multi-parameter coupling. At this stage, it faces the problems of low reliability and poor durability. Considering the correlation between operating life and cost, durability is one of the most important factors hindering the commercialization of PEMFC. The effective life prediction of PEMFC plays an important role in life durability research.

[0003] The remaining effective life prediction of PEMFC can predict the attenuation trend of PEMFC, improve the durability of PEMFC through energy management and other measures, and realize the long-term operation of PEMFC. Therefore, choosing a reasonable remaining effective life prediction method is crucial for the future industrial development of PEMFC. With the development of artificial intelligence and the advancement of computer technology, data-driven methods have received widespread attention in the field of PEMFC remaining effective life prediction. This method does not require the establishment of an accurate model, but directly uses the historical data of PEMFC stack operation for training to obtain the corresponding intelligent algorithm model, and predicts the PEMFC stack voltage based on the current operation data, which can effectively improve the prediction accuracy.

[0004] In the prior art, the prediction of the effective life of PEMFC is divided into three main types, namely, model-based methods, data-driven methods, and hybrid methods, depending on whether there is a mathematical model or an analytical model. Model-based methods have been widely studied due to their intuitiveness and revealing physical mechanisms, and some typical PEMFC remaining effective life prediction models have been explored. Hybrid methods are usually based on models and filtering or machine learning, but still require mathematical or analytical PEMFC related models. Data-driven methods predict the future decay state of PEMFC based on historical decay trends, do not require precise mathematical models or analytical models, and are suitable for constructing high-order nonlinear relationships between input and output signals. The data-driven remaining effective life prediction method mainly includes three steps: data processing, model training, and model prediction. The current innovation direction focuses on the improvement of deep learning algorithms to achieve accurate prediction. Deep learning methods currently widely used for remaining effective life prediction include adaptive network-based fuzzy inference system (ANFIS), long short-term memory (LSTM) neural network, echo state network (ESN) and other methods.

[0005] After analysis, the existing PEMFC remaining useful life prediction methods have the following main defects: the prediction model can reveal some physical mechanisms, but it is difficult to establish an accurate static or dynamic PEMFC mathematical model, which limits the overall performance of model-based and hybrid methods. The existing data-driven methods still have room for improvement in prediction accuracy, and the structure of some methods is too complex, which to some extent limits the application of prediction methods. Summary of the invention

[0006] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a proton exchange membrane fuel cell life prediction method based on deep learning. The method comprises the following steps:

[0007] Discrete wavelet transform is used to reduce noise of proton exchange membrane fuel cell data and extract feature data;

[0008] The characteristic data is input into a trained deep learning model to obtain predicted voltage data of a proton exchange membrane fuel cell, and then the remaining effective life of the battery is determined based on the predicted voltage data, wherein the deep learning model is constructed based on a Non-stationary Transformer model.

[0009] In one embodiment, the deep learning model includes an encoder and a decoder, the encoder includes a de-stationary attention module and a first feedforward network, the decoder includes a masked multi-head attention mechanism module, a multi-head attention mechanism module and a second feedforward network, the de-stationary attention module is used to approximate non-stationary information from stationary data, and the non-stationary information reflects the fluctuation of the data, and the masked multi-head attention mechanism module is used to mask the self-attention sublayer.

[0010] Compared with the prior art, the advantage of the present invention is that, firstly, the raw data is processed by a method based on discrete wavelet transform (DWT), and then the deep learning Non-stationary Transformer neural network algorithm is applied to the processed data to predict the remaining effective life of PEMFC, thereby effectively improving the prediction accuracy of the remaining effective life, predicting the attenuation state of PEMFC, facilitating the development of relevant energy management strategies, and ultimately improving the operating life of PEMFC.

[0011] Further features and advantages of the present invention will become apparent from the following detailed description of exemplary embodiments of the present invention with reference to the attached drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the invention and, together with the description, serve to explain the principles of the invention.

[0013] Figure 1 is a flow chart of a proton exchange membrane fuel cell life prediction method based on deep learning according to an embodiment of the present invention;

[0014] Figure 2 is a schematic diagram of a Non-stationary Transformer neural network framework according to an embodiment of the present invention;

[0015] Figure 3 The present invention is a flowchart of a PEMFC remaining useful life prediction strategy based on a Non-stationary Transformer neural network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0016] Various exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It should be noted that the relative arrangement of components and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present invention unless otherwise specifically stated.

[0017] The following description of at least one exemplary embodiment is merely illustrative in nature and is in no way intended to limit the invention, its application, or uses.

[0018] Technologies, methods, and equipment known to ordinary technicians in the relevant art may not be discussed in detail, but where appropriate, the technologies, methods, and equipment should be considered as part of the specification.

[0019] In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not limiting. Therefore, other examples of the exemplary embodiments may have different values.

[0020] It should be noted that like reference numerals and letters refer to similar items in the following figures, and therefore, once an item is defined in one figure, it need not be further discussed in subsequent figures.

[0021] In general, the deep learning-based proton exchange membrane fuel cell life prediction method provided by the present invention includes two parts. The first part includes the processing of PEMFC operation data, such as processing the original data using the DWT method; the second part is the prediction process of the PEMFC remaining effective life prediction method based on the Non-stationary Transformer algorithm, such as dividing the standardized data into a training set, a validation set and a test set for the implementation of the Non-stationary Transformer neural network remaining effective life prediction strategy, the training and validation process obtains a prediction model based on the Non-stationary Transformer, and the testing process or the model application process is the battery remaining effective life prediction process.

[0022] Specifically, see Figure 1 As shown, the provided proton exchange membrane fuel cell life prediction method based on deep learning includes the following steps:

[0023] Step S110, performing noise reduction processing on the PEMFC operation data using discrete wavelet transform.

[0024] DWT (discrete wavelet transform) is used to process the original data. DWT decomposes the original signal into discontinuous wavelets. By analyzing the changes of the signal in different time scales and frequency ranges, the local characteristics of the signal are extracted and the influence of noise is reduced. Since there is more information in time and frequency, it is better than Fourier transform in time resolution. The wavelet basis function is defined as:

[0025]

[0026] Among them, a is the scale factor, which is used to control the contraction range of the wavelet basis function, b is the translation factor, and t represents the time. Assuming a finite time series signal x(t), the continuous wavelet transform is defined as:

[0027]

[0028] The a and b of the wavelet transform are discrete points, and the result of the wavelet transform is obtained. For example, if a=2 m , b=n2 m , m,n∈Z, then the continuous wavelet is transformed into:

[0029]

[0030] The calculation method of wavelet basis function can be expressed as:

[0031]

[0032] After setting the detail coefficients to zero, the inverse transform W x (m,n) is converted back to the denoised data x'(t), and the calculation formula is expressed as:

[0033]

[0034] Data processed by DWT can effectively reduce noise in the data, improve data quality, and make subsequent analysis and prediction more accurate.

[0035] Step S120, building a deep learning model based on the Non-stationary Transformer model for predicting the remaining effective life of the battery.

[0036] Transformer is a model that uses the self-attention mechanism to increase the training speed. The training process of its basic units is parallel computing, which is more suitable for training in large-scale distributed clusters. When performing time series prediction, the model can capture the global dependency pattern of the original time series signal through the self-attention mechanism, and the diagnostic accuracy and generalization performance are significantly improved. However, the prediction process of Transformer is often accompanied by smooth transitions, resulting in similar information being learned for sequences with different attribute characteristics. On this basis, the Non-stationary Transformer algorithm improves the Transformer algorithm to alleviate the problem of over-stationarity in the prediction process.

[0037] 1) Transformer principle

[0038] The Transformer model is mainly composed of encoder and decoder modules. It adopts a sequence-to-sequence model structure. It divides the model input into k non-overlapping time series according to the time dimension and constructs a one-dimensional vector, which is linearly mapped to the model output. The encoder is composed of N identical small encoders. Each small encoder contains two layers of self-attention and feedforward neural network. After the data is processed by the self-attention module, it is input into the fully connected feedforward neural network. The decoder is also composed of N identical small decoders. The small decoder contains two self-attention layers and one feedforward neural network layer, and the first self-attention sublayer is masked (or called a masked multi-head attention layer), ensuring that the prediction of position i can only depend on the output of position less than i.

[0039] The self-attention layer uses the query-key-value (QKV) model, and its function can be described as mapping a Q vector and a set of (K, V) vector pairs to an output vector. k The Q vector and K vector of dimension and d v The output is calculated as the weighted sum of V vectors. The weight of each V vector is calculated by the compatibility function of the Q vector and the corresponding K vector. In fact, the self-attention function is to package a group of Q vectors into a matrix Q for calculation at the same time, and package the K vector and V vector into matrices K and V. Its output is calculated as follows:

[0040]

[0041] Among them, softmax(·) is the normalization function, is a scaling factor to prevent the input value of the normalization function from being too large, causing the partial derivative to approach zero.

[0042] 2) Non-stationary Transformer Principle

[0043] The Non-stationary Transformer algorithm adds a de-stationary attention module to the encoder, which can approximate non-stationary information from stationary data and reintroduce non-stationary information into the calculation of non-stationary sequences. This module contains learnable mean and variance, which can effectively alleviate the problem of over-stationarity. The framework of the Non-stationary Transformer algorithm is as follows: Figure 2 As shown. The transformation of using stationary data to approximate non-stationary attention is as follows:

[0044]

[0045] in, μ X is the mean of the input sequence X, σ Xis the variance of the input sequence X, μ Q , μ K and μ V They represent the average values ​​of vector Q, vector K and vector V in the time dimension respectively. From formula (7), we can see that the information of the original sequence not only contains the stable Q′ and K′ information, but also contains the σ of the non-stationary sequence. X , μ X and K information. The destabilization factors τ and Δ are used for the destabilization attention of each layer, respectively, from the statistics σ X , μ X and unstabilized X, which is calculated as follows:

[0046] logτ=MLP(σ X , X), Δ=MLP(μ X ,X) (8)

[0047] Finally, the output of the destabilizing attention module is:

[0048]

[0049] 3) Construction of remaining effective life prediction model based on Non-stationary Transformer.

[0050] Combination Figure 2 As shown in Figure 1, the deep learning model based on Non-stationary Transformer generally includes an encoder and a decoder. The encoder includes a de-stationary attention module, an addition and normalization layer, a feedforward network, etc. The decoder includes a masked multi-head attention mechanism layer, a multi-head attention mechanism layer, an addition and normalization layer, and a feedforward network, etc. Preferably, the input time series data X is normalized. X represents the characteristic data of PEMFC, and the normalized data is marked as X′.

[0051] Step S130, using the denoised data set to train the constructed deep learning model.

[0052] In this step, the constructed deep learning model is trained using the dataset until the set model accuracy requirements are met.

[0053] First, all the original features and label data provided by the data set are subjected to DWT denoising operations in turn to improve the quality and reliability of the data. The feature data and label data can be composed of aging parameters obtained in the experiment. The feature data include current, inlet and outlet temperature and pressure of the reaction gas, cooling water flow rate, cooling wind speed, etc. The label data is the PEMFC stack voltage. Then, the processed data are normalized to eliminate the numerical differences between them to reduce the impact on the prediction results. Then, the normalized data are divided into a training data set and a test data set, and a part of the training data set is used as a validation data set. Then, the training data set is used to train the deep learning model built based on the Non-stationary Transformer model, and the validation data set is used to validate it to evaluate the training effect, and the hyperparameters are optimized during the validation process. Under the premise that the prediction model structure and topological details are given, different combinations of hyperparameters may lead to different training results. Furthermore, the trained prediction model can be applied to the test data set, and the prediction results can be comprehensively evaluated until the model meets the accuracy requirements.

[0054] Step S140: predicting the remaining effective life of the proton exchange membrane fuel cell using the trained deep learning model.

[0055] Once the deep learning model is trained, it can be used to predict the remaining useful life of a proton exchange membrane fuel cell. Figure 3 As shown, the model application process generally includes: using discrete wavelet transform to reduce noise on the target PEM fuel cell data and extracting feature data; inputting the feature data into the trained deep learning model to obtain the predicted voltage data of the PEM fuel cell; according to the corresponding relationship between voltage and remaining battery life, the remaining effective life of the PEMFC can be obtained. Preferably, the feature data input to the model is normalized data.

[0056] In summary, the present invention first uses discrete wavelet transform DWT to perform noise reduction on the original data, and then develops a method based on deep learning Non-stationary Transformer neural network to predict the remaining effective life of PEMFC. By using relevant data to train the Non-stationary Transformer neural network, the trained prediction model can output the prediction result of the remaining effective life of the battery and improve the prediction accuracy. It has been verified that the present invention is suitable for the prediction of nonlinear systems, can make more accurate predictions on the remaining effective life of PEMFC, and can be extended to the life prediction of similar nonlinear systems.

[0057] It should be noted that the training process of the deep learning model involved in the present invention can be performed offline on a server or in the cloud, and the trained model can be embedded in an electronic device to achieve real-time prediction of the remaining battery life. The electronic device can be a terminal device or a server, and the terminal device includes any terminal device such as a mobile phone, a tablet computer, a personal digital assistant (PDA), a car computer, a smart wearable device, etc. The server includes but is not limited to an application server or a Web server, and can be an independent server or a cluster server or a cloud server, etc.

[0058] The present invention may be a system, a method and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for causing a processor to implement various aspects of the present invention.

[0059] Computer readable storage medium can be a tangible device that can hold and store instructions used by an instruction execution device. Computer readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination thereof. More specific examples (non-exhaustive list) of computer readable storage medium include: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a static random access memory (SRAM), a portable compact disk read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanical encoding device, for example, a punch card or a convex structure in a groove on which instructions are stored, and any suitable combination thereof. The computer readable storage medium used here is not interpreted as a transient signal itself, such as a radio wave or other freely propagating electromagnetic wave, an electromagnetic wave propagated by a waveguide or other transmission medium (for example, a light pulse by an optical fiber cable), or an electrical signal transmitted by a wire.

[0060] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to each computing / processing device, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and / or a wireless network. The network can include copper transmission cables, optical fiber transmissions, wireless transmissions, routers, firewalls, switches, gateway computers, and / or edge servers. The network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing / processing device.

[0061] The computer program instructions for performing the operation of the present invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages, such as Smalltalk, C++, Python, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. Computer-readable program instructions may be executed entirely on a user's computer, partially on a user's computer, as an independent software package, partially on a user's computer, partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., using an Internet service provider to connect via the Internet). In some embodiments, an electronic circuit, such as a programmable logic circuit, a field programmable gate array (FPGA), or a programmable logic array (PLA), may be personalized by utilizing the state information of the computer-readable program instructions, and the electronic circuit may execute the computer-readable program instructions, thereby realizing various aspects of the present invention.

[0062] Various aspects of the present invention are described herein with reference to the flow charts and / or block diagrams of the methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each box of the flow chart and / or block diagram and the combination of each box in the flow chart and / or block diagram can be implemented by computer-readable program instructions.

[0063] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processor of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0064] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operating steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0065] The flow charts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the systems, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a part of a module, a program segment or an instruction, and a part of the module, a program segment or an instruction contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented by a dedicated hardware-based system that performs the specified function or action, or can be implemented by a combination of dedicated hardware and computer instructions. It is well known to those skilled in the art that it is equivalent to implement it by hardware, implement it by software, and implement it by combining software and hardware.

[0066] Embodiments of the present invention have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or technical improvements in the marketplace, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein. The scope of the present invention is defined by the appended claims.

Claims

1. A proton exchange membrane fuel cell life prediction method based on deep learning, comprising the following steps: Discrete wavelet transform is used to reduce the noise of proton exchange membrane fuel cell data and extract feature data; The characteristic data is input into a trained deep learning model to obtain predicted voltage data of the proton exchange membrane fuel cell, and then the remaining effective life of the proton exchange membrane fuel cell is determined based on the predicted voltage data, wherein the deep learning model is constructed based on a Non-stationary Transformer model.

2. The method according to claim 1, characterized in that The deep learning model includes an encoder and a decoder, the encoder includes a de-stationary attention module and a first feedforward network, the decoder includes a masked multi-head attention mechanism module, a multi-head attention mechanism module and a second feedforward network, the de-stationary attention module is used to approximate non-stationary information from stationary data, and the non-stationary information reflects the fluctuation of the data, and the masked multi-head attention mechanism module is used to mask the self-attention sublayer.

3. The method according to claim 2, characterized in that The output of the destabilizing attention module is expressed as: in: Among them, τ is derived from the statistic σ X The destabilization factor learned in , Δ is the statistic μ X The destabilization factors learned in are expressed as: logτ=MLP(σ X ,X) Δ=MLP(μ X ,X) Where X represents the input sequence, σ X is the variance of X, μ X is the mean of X, Q, K, and V are the query vector, key vector, and value vector in the self-attention mechanism, respectively. Q' is the stabilized query matrix, K' is the stabilized key matrix, and V' is the stabilized value matrix. is the scaling factor, μ Q , μ K and μ V They represent the average values ​​of vectors Q, K and V in the time dimension respectively.

4. The method according to claim 3, characterized in that The self-attention mechanism is expressed as: Among them, softmax(·) is the normalization function.

5. The method according to claim 1, characterized in that The characteristic data includes one or more of the current, inlet and outlet temperatures and pressures of the reaction gas, cooling water flow rate, and cooling wind speed.

6. The method according to claim 1, characterized in that The deep learning model is trained on a server or in the cloud.

7. The method according to claim 1, characterized in that The feature data input into the deep learning model is normalized feature data.

8. The method according to claim 1, characterized in that: The remaining effective life of the proton exchange membrane fuel cell is determined based on a nonlinear relationship between the predicted voltage data and the remaining life of the proton exchange membrane fuel cell.

9. A computer-readable storage medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer device comprising a memory and a processor, wherein a computer program capable of running on the processor is stored in the memory, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images